晶圓廠小批量之報廢準則研究

Abstract

半導體廠一晶圓批的批量通常是25片，有些晶片因加工不良變成廢品，因此在製造過程中，一個晶圓批的批量可能會少於25片；吾人稱此種少於25片的晶圓批為「小量晶圓批」(small wafer lot)。 小量晶圓批是否應該繼續加工或逕行報廢，是晶圓廠經常面臨的一個決策問題。但是在過去的學術文獻上，尚未發現有相關的研究。工業界處理這個問題通常是依據以直覺方式處理，缺乏有系統的分析方法。本研究計畫擬探討小量晶圓批報廢的決策問題，以基因演算法，來決定小量晶圓批在各加工道次的報廢法則，期使晶圓廠的利潤最大化。

In semiconductor manufacturing, a typical wafer lot is composed of 25 wafers. Due to the yield problems, wafers in a lot may decrease in number during manufacturing. Such a lot is called a small wafer lot. The unit cost of manufacturing the wafers in a small lot is more expensive than a normal lot. Yet, completing the manufacture of a small lot would create revenue. Semiconductor fab. therefore always have to make the decision—under what conditions should a small wafer lot be scraped. According to our survey, no literature has been published on this research problem. And the decision made in industry is often by the intuition of shop managers. The main objective of this paper aims to develop the decision-making methods for scrapping small lots in semiconductor fab. in order to maximize the profit. We intend to develop genetic algorithm methods under a simplified production scenario. One main assumption is that the bottleneck machine is a particular type.